/*

 * Copyright (c) 2018, 2019, Red Hat, Inc. All rights reserved.

 *

 * 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 "gc/shared/barrierSet.hpp"

#include "gc/shenandoah/shenandoahForwarding.hpp"

#include "gc/shenandoah/shenandoahHeap.hpp"

#include "gc/shenandoah/shenandoahHeuristics.hpp"

#include "gc/shenandoah/shenandoahRuntime.hpp"

#include "gc/shenandoah/shenandoahThreadLocalData.hpp"

#include "gc/shenandoah/c2/shenandoahBarrierSetC2.hpp"

#include "gc/shenandoah/c2/shenandoahSupport.hpp"

#include "opto/arraycopynode.hpp"

#include "opto/escape.hpp"

#include "opto/graphKit.hpp"

#include "opto/idealKit.hpp"

#include "opto/macro.hpp"

#include "opto/movenode.hpp"

#include "opto/narrowptrnode.hpp"

#include "opto/rootnode.hpp"

#include "opto/runtime.hpp"

ShenandoahBarrierSetC2* ShenandoahBarrierSetC2::bsc2() {

  return reinterpret_cast<ShenandoahBarrierSetC2*>(BarrierSet::barrier_set()->barrier_set_c2());

}

ShenandoahBarrierSetC2State::ShenandoahBarrierSetC2State(Arena* comp_arena)

  : _enqueue_barriers(new (comp_arena) GrowableArray<ShenandoahEnqueueBarrierNode*>(comp_arena, 8,  0, NULL)),

    _load_reference_barriers(new (comp_arena) GrowableArray<ShenandoahLoadReferenceBarrierNode*>(comp_arena, 8,  0, NULL)) {

}

int ShenandoahBarrierSetC2State::enqueue_barriers_count() const {

  return _enqueue_barriers->length();

}

ShenandoahEnqueueBarrierNode* ShenandoahBarrierSetC2State::enqueue_barrier(int idx) const {

  return _enqueue_barriers->at(idx);

}

void ShenandoahBarrierSetC2State::add_enqueue_barrier(ShenandoahEnqueueBarrierNode * n) {

  assert(!_enqueue_barriers->contains(n), "duplicate entry in barrier list");

  _enqueue_barriers->append(n);

}

void ShenandoahBarrierSetC2State::remove_enqueue_barrier(ShenandoahEnqueueBarrierNode * n) {

  if (_enqueue_barriers->contains(n)) {

    _enqueue_barriers->remove(n);

  }

}

int ShenandoahBarrierSetC2State::load_reference_barriers_count() const {

  return _load_reference_barriers->length();

}

ShenandoahLoadReferenceBarrierNode* ShenandoahBarrierSetC2State::load_reference_barrier(int idx) const {

  return _load_reference_barriers->at(idx);

}

void ShenandoahBarrierSetC2State::add_load_reference_barrier(ShenandoahLoadReferenceBarrierNode * n) {

  assert(!_load_reference_barriers->contains(n), "duplicate entry in barrier list");

  _load_reference_barriers->append(n);

}

void ShenandoahBarrierSetC2State::remove_load_reference_barrier(ShenandoahLoadReferenceBarrierNode * n) {

  if (_load_reference_barriers->contains(n)) {

    _load_reference_barriers->remove(n);

  }

}

Node* ShenandoahBarrierSetC2::shenandoah_storeval_barrier(GraphKit* kit, Node* obj) const {

  if (ShenandoahStoreValEnqueueBarrier) {

    obj = shenandoah_enqueue_barrier(kit, obj);

  }

  return obj;

}

#define __ kit->

bool ShenandoahBarrierSetC2::satb_can_remove_pre_barrier(GraphKit* kit, PhaseTransform* phase, Node* adr,

                                                         BasicType bt, uint adr_idx) const {

  intptr_t offset = 0;

  Node* base = AddPNode::Ideal_base_and_offset(adr, phase, offset);

  AllocateNode* alloc = AllocateNode::Ideal_allocation(base, phase);

  if (offset == Type::OffsetBot) {

    return false; // cannot unalias unless there are precise offsets

  }

  if (alloc == NULL) {

    return false; // No allocation found

  }

  intptr_t size_in_bytes = type2aelembytes(bt);

  Node* mem = __ memory(adr_idx); // start searching here...

  for (int cnt = 0; cnt < 50; cnt++) {

    if (mem->is_Store()) {

      Node* st_adr = mem->in(MemNode::Address);

      intptr_t st_offset = 0;

      Node* st_base = AddPNode::Ideal_base_and_offset(st_adr, phase, st_offset);

      if (st_base == NULL) {

        break; // inscrutable pointer

      }

      // Break we have found a store with same base and offset as ours so break

      if (st_base == base && st_offset == offset) {

        break;

      }

      if (st_offset != offset && st_offset != Type::OffsetBot) {

        const int MAX_STORE = BytesPerLong;

        if (st_offset >= offset + size_in_bytes ||

            st_offset <= offset - MAX_STORE ||

            st_offset <= offset - mem->as_Store()->memory_size()) {

          // Success:  The offsets are provably independent.

          // (You may ask, why not just test st_offset != offset and be done?

          // The answer is that stores of different sizes can co-exist

          // in the same sequence of RawMem effects.  We sometimes initialize

          // a whole 'tile' of array elements with a single jint or jlong.)

          mem = mem->in(MemNode::Memory);

          continue; // advance through independent store memory

        }

      }

      if (st_base != base

          && MemNode::detect_ptr_independence(base, alloc, st_base,

                                              AllocateNode::Ideal_allocation(st_base, phase),

                                              phase)) {

        // Success:  The bases are provably independent.

        mem = mem->in(MemNode::Memory);

        continue; // advance through independent store memory

      }

    } else if (mem->is_Proj() && mem->in(0)->is_Initialize()) {

      InitializeNode* st_init = mem->in(0)->as_Initialize();

      AllocateNode* st_alloc = st_init->allocation();

      // Make sure that we are looking at the same allocation site.

      // The alloc variable is guaranteed to not be null here from earlier check.

      if (alloc == st_alloc) {

        // Check that the initialization is storing NULL so that no previous store

        // has been moved up and directly write a reference

        Node* captured_store = st_init->find_captured_store(offset,

                                                            type2aelembytes(T_OBJECT),

                                                            phase);

        if (captured_store == NULL || captured_store == st_init->zero_memory()) {

          return true;

        }

      }

    }

    // Unless there is an explicit 'continue', we must bail out here,

    // because 'mem' is an inscrutable memory state (e.g., a call).

    break;

  }

  return false;

}

#undef __

#define __ ideal.

void ShenandoahBarrierSetC2::satb_write_barrier_pre(GraphKit* kit,

                                                    bool do_load,

                                                    Node* obj,

                                                    Node* adr,

                                                    uint alias_idx,

                                                    Node* val,

                                                    const TypeOopPtr* val_type,

                                                    Node* pre_val,

                                                    BasicType bt) const {

  // Some sanity checks

  // Note: val is unused in this routine.

  if (do_load) {

    // We need to generate the load of the previous value

    assert(obj != NULL, "must have a base");

    assert(adr != NULL, "where are loading from?");

    assert(pre_val == NULL, "loaded already?");

    assert(val_type != NULL, "need a type");

    if (ReduceInitialCardMarks

        && satb_can_remove_pre_barrier(kit, &kit->gvn(), adr, bt, alias_idx)) {

      return;

    }

  } else {

    // In this case both val_type and alias_idx are unused.

    assert(pre_val != NULL, "must be loaded already");

    // Nothing to be done if pre_val is null.

    if (pre_val->bottom_type() == TypePtr::NULL_PTR) return;

    assert(pre_val->bottom_type()->basic_type() == T_OBJECT, "or we shouldn't be here");

  }

  assert(bt == T_OBJECT, "or we shouldn't be here");

  IdealKit ideal(kit, true);

  Node* tls = __ thread(); // ThreadLocalStorage

  Node* no_base = __ top();

  Node* zero  = __ ConI(0);

  Node* zeroX = __ ConX(0);

  float likely  = PROB_LIKELY(0.999);

  float unlikely  = PROB_UNLIKELY(0.999);

  // Offsets into the thread

  const int index_offset   = in_bytes(ShenandoahThreadLocalData::satb_mark_queue_index_offset());

  const int buffer_offset  = in_bytes(ShenandoahThreadLocalData::satb_mark_queue_buffer_offset());

  // Now the actual pointers into the thread

  Node* buffer_adr  = __ AddP(no_base, tls, __ ConX(buffer_offset));

  Node* index_adr   = __ AddP(no_base, tls, __ ConX(index_offset));

  // Now some of the values

  Node* marking;

  Node* gc_state = __ AddP(no_base, tls, __ ConX(in_bytes(ShenandoahThreadLocalData::gc_state_offset())));

  Node* ld = __ load(__ ctrl(), gc_state, TypeInt::BYTE, T_BYTE, Compile::AliasIdxRaw);

  marking = __ AndI(ld, __ ConI(ShenandoahHeap::MARKING));

  assert(ShenandoahBarrierC2Support::is_gc_state_load(ld), "Should match the shape");

  // if (!marking)

  __ if_then(marking, BoolTest::ne, zero, unlikely); {

    BasicType index_bt = TypeX_X->basic_type();

    assert(sizeof(size_t) == type2aelembytes(index_bt), "Loading G1 SATBMarkQueue::_index with wrong size.");

    Node* index   = __ load(__ ctrl(), index_adr, TypeX_X, index_bt, Compile::AliasIdxRaw);

    if (do_load) {

      // load original value

      // alias_idx correct??

      pre_val = __ load(__ ctrl(), adr, val_type, bt, alias_idx);

    }

    // if (pre_val != NULL)

    __ if_then(pre_val, BoolTest::ne, kit->null()); {

      Node* buffer  = __ load(__ ctrl(), buffer_adr, TypeRawPtr::NOTNULL, T_ADDRESS, Compile::AliasIdxRaw);

      // is the queue for this thread full?

      __ if_then(index, BoolTest::ne, zeroX, likely); {

        // decrement the index

        Node* next_index = kit->gvn().transform(new SubXNode(index, __ ConX(sizeof(intptr_t))));

        // Now get the buffer location we will log the previous value into and store it

        Node *log_addr = __ AddP(no_base, buffer, next_index);

        __ store(__ ctrl(), log_addr, pre_val, T_OBJECT, Compile::AliasIdxRaw, MemNode::unordered);

        // update the index

        __ store(__ ctrl(), index_adr, next_index, index_bt, Compile::AliasIdxRaw, MemNode::unordered);

      } __ else_(); {

        // logging buffer is full, call the runtime

        const TypeFunc *tf = ShenandoahBarrierSetC2::write_ref_field_pre_entry_Type();

        __ make_leaf_call(tf, CAST_FROM_FN_PTR(address, ShenandoahRuntime::write_ref_field_pre_entry), "shenandoah_wb_pre", pre_val, tls);

      } __ end_if();  // (!index)

    } __ end_if();  // (pre_val != NULL)

  } __ end_if();  // (!marking)

  // Final sync IdealKit and GraphKit.

  kit->final_sync(ideal);

  if (ShenandoahSATBBarrier && adr != NULL) {

    Node* c = kit->control();

    Node* call = c->in(1)->in(1)->in(1)->in(0);

    assert(is_shenandoah_wb_pre_call(call), "shenandoah_wb_pre call expected");

    call->add_req(adr);

  }

}

bool ShenandoahBarrierSetC2::is_shenandoah_wb_pre_call(Node* call) {

  return call->is_CallLeaf() &&

         call->as_CallLeaf()->entry_point() == CAST_FROM_FN_PTR(address, ShenandoahRuntime::write_ref_field_pre_entry);

}

bool ShenandoahBarrierSetC2::is_shenandoah_lrb_call(Node* call) {

}

bool ShenandoahBarrierSetC2::is_shenandoah_marking_if(PhaseTransform *phase, Node* n) {

  if (n->Opcode() != Op_If) {

    return false;

  }

  Node* bol = n->in(1);

  assert(bol->is_Bool(), "");

  Node* cmpx = bol->in(1);

  if (bol->as_Bool()->_test._test == BoolTest::ne &&

      cmpx->is_Cmp() && cmpx->in(2) == phase->intcon(0) &&

      is_shenandoah_state_load(cmpx->in(1)->in(1)) &&

      cmpx->in(1)->in(2)->is_Con() &&

      cmpx->in(1)->in(2) == phase->intcon(ShenandoahHeap::MARKING)) {

    return true;

  }

  return false;

}

bool ShenandoahBarrierSetC2::is_shenandoah_state_load(Node* n) {

  if (!n->is_Load()) return false;

  const int state_offset = in_bytes(ShenandoahThreadLocalData::gc_state_offset());

  return n->in(2)->is_AddP() && n->in(2)->in(2)->Opcode() == Op_ThreadLocal

         && n->in(2)->in(3)->is_Con()

         && n->in(2)->in(3)->bottom_type()->is_intptr_t()->get_con() == state_offset;

}

void ShenandoahBarrierSetC2::shenandoah_write_barrier_pre(GraphKit* kit,

                                                          bool do_load,

                                                          Node* obj,

                                                          Node* adr,

                                                          uint alias_idx,

                                                          Node* val,

                                                          const TypeOopPtr* val_type,

                                                          Node* pre_val,

                                                          BasicType bt) const {

  if (ShenandoahSATBBarrier) {

    IdealKit ideal(kit);

    kit->sync_kit(ideal);

    satb_write_barrier_pre(kit, do_load, obj, adr, alias_idx, val, val_type, pre_val, bt);

    ideal.sync_kit(kit);

    kit->final_sync(ideal);

  }

}

Node* ShenandoahBarrierSetC2::shenandoah_enqueue_barrier(GraphKit* kit, Node* pre_val) const {

  return kit->gvn().transform(new ShenandoahEnqueueBarrierNode(pre_val));

}

// Helper that guards and inserts a pre-barrier.

void ShenandoahBarrierSetC2::insert_pre_barrier(GraphKit* kit, Node* base_oop, Node* offset,

                                                Node* pre_val, bool need_mem_bar) const {

  // We could be accessing the referent field of a reference object. If so, when G1

  // is enabled, we need to log the value in the referent field in an SATB buffer.

  // This routine performs some compile time filters and generates suitable

  // runtime filters that guard the pre-barrier code.

  // Also add memory barrier for non volatile load from the referent field

  // to prevent commoning of loads across safepoint.

  // Some compile time checks.

  // If offset is a constant, is it java_lang_ref_Reference::_reference_offset?

  const TypeX* otype = offset->find_intptr_t_type();

  if (otype != NULL && otype->is_con() &&

      otype->get_con() != java_lang_ref_Reference::referent_offset) {

    // Constant offset but not the reference_offset so just return

    return;

  }

  // We only need to generate the runtime guards for instances.

  const TypeOopPtr* btype = base_oop->bottom_type()->isa_oopptr();

  if (btype != NULL) {

    if (btype->isa_aryptr()) {

      // Array type so nothing to do

      return;

    }

    const TypeInstPtr* itype = btype->isa_instptr();

    if (itype != NULL) {

      // Can the klass of base_oop be statically determined to be

      // _not_ a sub-class of Reference and _not_ Object?

      ciKlass* klass = itype->klass();

      if ( klass->is_loaded() &&

          !klass->is_subtype_of(kit->env()->Reference_klass()) &&

          !kit->env()->Object_klass()->is_subtype_of(klass)) {

        return;

      }

    }

  }

  // The compile time filters did not reject base_oop/offset so

  // we need to generate the following runtime filters

  //

  // if (offset == java_lang_ref_Reference::_reference_offset) {

  //   if (instance_of(base, java.lang.ref.Reference)) {

  //     pre_barrier(_, pre_val, ...);

  //   }

  // }

  float likely   = PROB_LIKELY(  0.999);

  float unlikely = PROB_UNLIKELY(0.999);

  IdealKit ideal(kit);

  Node* referent_off = __ ConX(java_lang_ref_Reference::referent_offset);

  __ if_then(offset, BoolTest::eq, referent_off, unlikely); {

      // Update graphKit memory and control from IdealKit.

      kit->sync_kit(ideal);

      Node* ref_klass_con = kit->makecon(TypeKlassPtr::make(kit->env()->Reference_klass()));

      Node* is_instof = kit->gen_instanceof(base_oop, ref_klass_con);

      // Update IdealKit memory and control from graphKit.

      __ sync_kit(kit);

      Node* one = __ ConI(1);

      // is_instof == 0 if base_oop == NULL

      __ if_then(is_instof, BoolTest::eq, one, unlikely); {

        // Update graphKit from IdeakKit.

        kit->sync_kit(ideal);

        // Use the pre-barrier to record the value in the referent field

        satb_write_barrier_pre(kit, false /* do_load */,

                               NULL /* obj */, NULL /* adr */, max_juint /* alias_idx */, NULL /* val */, NULL /* val_type */,

                               pre_val /* pre_val */,

                               T_OBJECT);

        if (need_mem_bar) {

          // Add memory barrier to prevent commoning reads from this field

          // across safepoint since GC can change its value.

          kit->insert_mem_bar(Op_MemBarCPUOrder);

        }

        // Update IdealKit from graphKit.

        __ sync_kit(kit);

      } __ end_if(); // _ref_type != ref_none

  } __ end_if(); // offset == referent_offset

  // Final sync IdealKit and GraphKit.

  kit->final_sync(ideal);

}

#undef __

const TypeFunc* ShenandoahBarrierSetC2::write_ref_field_pre_entry_Type() {

  const Type **fields = TypeTuple::fields(2);

  fields[TypeFunc::Parms+0] = TypeInstPtr::NOTNULL; // original field value

  fields[TypeFunc::Parms+1] = TypeRawPtr::NOTNULL; // thread

  const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms+2, fields);

  // create result type (range)

  fields = TypeTuple::fields(0);

  const TypeTuple *range = TypeTuple::make(TypeFunc::Parms+0, fields);

  return TypeFunc::make(domain, range);

}

const TypeFunc* ShenandoahBarrierSetC2::shenandoah_clone_barrier_Type() {

  const Type **fields = TypeTuple::fields(1);

  fields[TypeFunc::Parms+0] = TypeOopPtr::NOTNULL; // src oop

  const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms+1, fields);

  // create result type (range)

  fields = TypeTuple::fields(0);

  const TypeTuple *range = TypeTuple::make(TypeFunc::Parms+0, fields);

  return TypeFunc::make(domain, range);

}

const TypeFunc* ShenandoahBarrierSetC2::shenandoah_load_reference_barrier_Type() {

  const Type **fields = TypeTuple::fields(2);

  fields[TypeFunc::Parms+0] = TypeInstPtr::NOTNULL; // original field value

  fields[TypeFunc::Parms+1] = TypeRawPtr::BOTTOM;   // original load address

  const TypeTuple *domain = TypeTuple::make(TypeFunc::Parms+2, fields);

  // create result type (range)

  fields = TypeTuple::fields(1);

  fields[TypeFunc::Parms+0] = TypeInstPtr::NOTNULL;

  const TypeTuple *range = TypeTuple::make(TypeFunc::Parms+1, fields);

  return TypeFunc::make(domain, range);

}

Node* ShenandoahBarrierSetC2::store_at_resolved(C2Access& access, C2AccessValue& val) const {

  DecoratorSet decorators = access.decorators();

  const TypePtr* adr_type = access.addr().type();

  Node* adr = access.addr().node();

  bool anonymous = (decorators & ON_UNKNOWN_OOP_REF) != 0;

  bool on_heap = (decorators & IN_HEAP) != 0;

  if (!access.is_oop() || (!on_heap && !anonymous)) {

    return BarrierSetC2::store_at_resolved(access, val);

  }

  if (access.is_parse_access()) {

    C2ParseAccess& parse_access = static_cast<C2ParseAccess&>(access);

    GraphKit* kit = parse_access.kit();

    uint adr_idx = kit->C->get_alias_index(adr_type);

    assert(adr_idx != Compile::AliasIdxTop, "use other store_to_memory factory" );

    Node* value = val.node();

    value = shenandoah_storeval_barrier(kit, value);

    val.set_node(value);

    shenandoah_write_barrier_pre(kit, true /* do_load */, /*kit->control(),*/ access.base(), adr, adr_idx, val.node(),

                                 static_cast<const TypeOopPtr*>(val.type()), NULL /* pre_val */, access.type());

  } else {

    assert(access.is_opt_access(), "only for optimization passes");