/*

 * Copyright (c) 2018, Oracle and/or its affiliates. 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 "gc/g1/c2/g1BarrierSetC2.hpp"

#include "gc/g1/g1BarrierSet.hpp"

#include "gc/g1/g1BarrierSetRuntime.hpp"

#include "gc/g1/g1CardTable.hpp"

#include "gc/g1/g1ThreadLocalData.hpp"

#include "gc/g1/heapRegion.hpp"

#include "opto/arraycopynode.hpp"

#include "opto/compile.hpp"

#include "opto/graphKit.hpp"

#include "opto/idealKit.hpp"

#include "opto/macro.hpp"

#include "opto/rootnode.hpp"

#include "opto/type.hpp"

#include "utilities/macros.hpp"

const TypeFunc *G1BarrierSetC2::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 *G1BarrierSetC2::write_ref_field_post_entry_Type() {

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

  fields[TypeFunc::Parms+0] = TypeRawPtr::NOTNULL;  // Card addr

  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, fields);

  return TypeFunc::make(domain, range);

}

#define __ ideal.

/*

 * Determine if the G1 pre-barrier can be removed. The pre-barrier is

 * required by SATB to make sure all objects live at the start of the

 * marking are kept alive, all reference updates need to any previous

 * reference stored before writing.

 *

 * If the previous value is NULL there is no need to save the old value.

 * References that are NULL are filtered during runtime by the barrier

 * code to avoid unnecessary queuing.

 *

 * However in the case of newly allocated objects it might be possible to

 * prove that the reference about to be overwritten is NULL during compile

 * time and avoid adding the barrier code completely.

 *

 * The compiler needs to determine that the object in which a field is about

 * to be written is newly allocated, and that no prior store to the same field

 * has happened since the allocation.

 *

 * Returns true if the pre-barrier can be removed

 */

bool G1BarrierSetC2::g1_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 = kit->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;

}

// G1 pre/post barriers

void G1BarrierSetC2::pre_barrier(GraphKit* kit,

                                 bool do_load,

                                 Node* ctl,

                                 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 (use_ReduceInitialCardMarks()

        && g1_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);

  BasicType active_type = in_bytes(SATBMarkQueue::byte_width_of_active()) == 4 ? T_INT : T_BYTE;

  assert(in_bytes(SATBMarkQueue::byte_width_of_active()) == 4 || in_bytes(SATBMarkQueue::byte_width_of_active()) == 1, "flag width");

  // Offsets into the thread

  const int marking_offset = in_bytes(G1ThreadLocalData::satb_mark_queue_active_offset());

  const int index_offset   = in_bytes(G1ThreadLocalData::satb_mark_queue_index_offset());

  const int buffer_offset  = in_bytes(G1ThreadLocalData::satb_mark_queue_buffer_offset());

  // Now the actual pointers into the thread

  Node* marking_adr = __ AddP(no_base, tls, __ ConX(marking_offset));

  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 = __ load(__ ctrl(), marking_adr, TypeInt::INT, active_type, Compile::AliasIdxRaw);

  // 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 = write_ref_field_pre_entry_Type();

        __ make_leaf_call(tf, CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_pre_entry), "write_ref_field_pre_entry", pre_val, tls);

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

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

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

  // Final sync IdealKit and GraphKit.

  kit->final_sync(ideal);

}

/*

 * G1 similar to any GC with a Young Generation requires a way to keep track of

 * references from Old Generation to Young Generation to make sure all live

 * objects are found. G1 also requires to keep track of object references

 * between different regions to enable evacuation of old regions, which is done

 * as part of mixed collections. References are tracked in remembered sets and

 * is continuously updated as reference are written to with the help of the

 * post-barrier.

 *

 * To reduce the number of updates to the remembered set the post-barrier

 * filters updates to fields in objects located in the Young Generation,

 * the same region as the reference, when the NULL is being written or

 * if the card is already marked as dirty by an earlier write.

 *

 * Under certain circumstances it is possible to avoid generating the

 * post-barrier completely if it is possible during compile time to prove

 * the object is newly allocated and that no safepoint exists between the

 * allocation and the store.

 *

 * In the case of slow allocation the allocation code must handle the barrier

 * as part of the allocation in the case the allocated object is not located

 * in the nursery, this would happen for humongous objects. This is similar to

 * how CMS is required to handle this case, see the comments for the method

 * CollectedHeap::new_deferred_store_barrier and OptoRuntime::new_deferred_store_barrier.

 * A deferred card mark is required for these objects and handled in the above

 * mentioned methods.

 *

 * Returns true if the post barrier can be removed

 */

bool G1BarrierSetC2::g1_can_remove_post_barrier(GraphKit* kit,

                                                PhaseTransform* phase, Node* store,

                                                Node* adr) 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

  }

  // Start search from Store node

  Node* mem = store->in(MemNode::Control);

  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 we are looking at the same allocation

    if (alloc == st_alloc) {

      return true;

    }

  }

  return false;

}

//

// Update the card table and add card address to the queue

//

void G1BarrierSetC2::g1_mark_card(GraphKit* kit,

                                  IdealKit& ideal,

                                  Node* card_adr,

                                  Node* oop_store,

                                  uint oop_alias_idx,

                                  Node* index,

                                  Node* index_adr,

                                  Node* buffer,

                                  const TypeFunc* tf) const {

  Node* zero  = __ ConI(0);

  Node* zeroX = __ ConX(0);

  Node* no_base = __ top();

  BasicType card_bt = T_BYTE;

  // Smash zero into card. MUST BE ORDERED WRT TO STORE

  __ storeCM(__ ctrl(), card_adr, zero, oop_store, oop_alias_idx, card_bt, Compile::AliasIdxRaw);

  //  Now do the queue work

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

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

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

    // Order, see storeCM.

    __ store(__ ctrl(), log_addr, card_adr, T_ADDRESS, Compile::AliasIdxRaw, MemNode::unordered);

    __ store(__ ctrl(), index_adr, next_index, TypeX_X->basic_type(), Compile::AliasIdxRaw, MemNode::unordered);

  } __ else_(); {

    __ make_leaf_call(tf, CAST_FROM_FN_PTR(address, G1BarrierSetRuntime::write_ref_field_post_entry), "write_ref_field_post_entry", card_adr, __ thread());

  } __ end_if();

}

void G1BarrierSetC2::post_barrier(GraphKit* kit,

                                  Node* ctl,

                                  Node* oop_store,

                                  Node* obj,

                                  Node* adr,

                                  uint alias_idx,

                                  Node* val,

                                  BasicType bt,

                                  bool use_precise) const {

  // If we are writing a NULL then we need no post barrier

  if (val != NULL && val->is_Con() && val->bottom_type() == TypePtr::NULL_PTR) {

    // Must be NULL

    const Type* t = val->bottom_type();

    assert(t == Type::TOP || t == TypePtr::NULL_PTR, "must be NULL");

    // No post barrier if writing NULLx

    return;

  }

  if (use_ReduceInitialCardMarks() && obj == kit->just_allocated_object(kit->control())) {

    // We can skip marks on a freshly-allocated object in Eden.

    // Keep this code in sync with new_deferred_store_barrier() in runtime.cpp.

    // That routine informs GC to take appropriate compensating steps,

    // upon a slow-path allocation, so as to make this card-mark

    // elision safe.

    return;

  }

  if (use_ReduceInitialCardMarks()

      && g1_can_remove_post_barrier(kit, &kit->gvn(), oop_store, adr)) {

    return;

  }

  if (!use_precise) {

    // All card marks for a (non-array) instance are in one place:

    adr = obj;

  }

  // (Else it's an array (or unknown), and we want more precise card marks.)

  assert(adr != NULL, "");

  IdealKit ideal(kit, true);

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

  Node* no_base = __ top();

  float unlikely  = PROB_UNLIKELY(0.999);

  Node* young_card = __ ConI((jint)G1CardTable::g1_young_card_val());

  Node* dirty_card = __ ConI((jint)G1CardTable::dirty_card_val());

  Node* zeroX = __ ConX(0);

  const TypeFunc *tf = write_ref_field_post_entry_Type();

  // Offsets into the thread

  const int index_offset  = in_bytes(G1ThreadLocalData::dirty_card_queue_index_offset());

  const int buffer_offset = in_bytes(G1ThreadLocalData::dirty_card_queue_buffer_offset());

  // 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 values

  // Use ctrl to avoid hoisting these values past a safepoint, which could

  // potentially reset these fields in the JavaThread.

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

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

  // Convert the store obj pointer to an int prior to doing math on it

  // Must use ctrl to prevent "integerized oop" existing across safepoint

  Node* cast =  __ CastPX(__ ctrl(), adr);

  // Divide pointer by card size

  Node* card_offset = __ URShiftX( cast, __ ConI(CardTable::card_shift) );

  // Combine card table base and card offset

  Node* card_adr = __ AddP(no_base, byte_map_base_node(kit), card_offset );

  // If we know the value being stored does it cross regions?

  if (val != NULL) {

    // Does the store cause us to cross regions?

    // Should be able to do an unsigned compare of region_size instead of

    // and extra shift. Do we have an unsigned compare??

    // Node* region_size = __ ConI(1 << HeapRegion::LogOfHRGrainBytes);

    Node* xor_res =  __ URShiftX ( __ XorX( cast,  __ CastPX(__ ctrl(), val)), __ ConI(HeapRegion::LogOfHRGrainBytes));

    // if (xor_res == 0) same region so skip

    __ if_then(xor_res, BoolTest::ne, zeroX); {

      // No barrier if we are storing a NULL

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

        // Ok must mark the card if not already dirty

        // load the original value of the card

        Node* card_val = __ load(__ ctrl(), card_adr, TypeInt::INT, T_BYTE, Compile::AliasIdxRaw);

        __ if_then(card_val, BoolTest::ne, young_card); {

          kit->sync_kit(ideal);

          kit->insert_mem_bar(Op_MemBarVolatile, oop_store);

          __ sync_kit(kit);

          Node* card_val_reload = __ load(__ ctrl(), card_adr, TypeInt::INT, T_BYTE, Compile::AliasIdxRaw);

          __ if_then(card_val_reload, BoolTest::ne, dirty_card); {

            g1_mark_card(kit, ideal, card_adr, oop_store, alias_idx, index, index_adr, buffer, tf);

          } __ end_if();

        } __ end_if();

      } __ end_if();

    } __ end_if();

  } else {

    // The Object.clone() intrinsic uses this path if !ReduceInitialCardMarks.

    // We don't need a barrier here if the destination is a newly allocated object

    // in Eden. Otherwise, GC verification breaks because we assume that cards in Eden

    // are set to 'g1_young_gen' (see G1CardTable::verify_g1_young_region()).

    assert(!use_ReduceInitialCardMarks(), "can only happen with card marking");

    Node* card_val = __ load(__ ctrl(), card_adr, TypeInt::INT, T_BYTE, Compile::AliasIdxRaw);

    __ if_then(card_val, BoolTest::ne, young_card); {

      g1_mark_card(kit, ideal, card_adr, oop_store, alias_idx, index, index_adr, buffer, tf);

    } __ end_if();

  }

  // Final sync IdealKit and GraphKit.

  kit->final_sync(ideal);

}

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

void G1BarrierSetC2::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;

    }