/*

 * Copyright (c) 2018, 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/shenandoah/shenandoahBarrierSetAssembler.hpp"

#include "gc/shenandoah/shenandoahHeap.hpp"

#include "gc/shenandoah/shenandoahHeapRegion.hpp"

#include "gc/shenandoah/shenandoahHeuristics.hpp"

#include "gc/shenandoah/shenandoahRuntime.hpp"

#include "gc/shenandoah/shenandoahThreadLocalData.hpp"

#include "interpreter/interpreter.hpp"

#include "interpreter/interp_masm.hpp"

#include "runtime/sharedRuntime.hpp"

#include "runtime/thread.hpp"

#include "utilities/macros.hpp"

#ifdef COMPILER1

#include "c1/c1_LIRAssembler.hpp"

#include "c1/c1_MacroAssembler.hpp"

#include "gc/shenandoah/c1/shenandoahBarrierSetC1.hpp"

#endif

#define __ masm->

void ShenandoahBarrierSetAssembler::arraycopy_prologue(MacroAssembler* masm, DecoratorSet decorators, BasicType type,

                                                       Register src, Register dst, Register count) {

  bool checkcast = (decorators & ARRAYCOPY_CHECKCAST) != 0;

  bool disjoint = (decorators & ARRAYCOPY_DISJOINT) != 0;

  bool obj_int = type == T_OBJECT LP64_ONLY(&& UseCompressedOops);

  bool dest_uninitialized = (decorators & IS_DEST_UNINITIALIZED) != 0;

  if (type == T_OBJECT || type == T_ARRAY) {

#ifdef _LP64

    if (!checkcast && !obj_int) {

      // Save count for barrier

      __ movptr(r11, count);

    } else if (disjoint && obj_int) {

      // Save dst in r11 in the disjoint case

      __ movq(r11, dst);

    }

#else

    if (disjoint) {

      __ mov(rdx, dst);          // save 'to'

    }

#endif

    if (ShenandoahSATBBarrier && !dest_uninitialized && !ShenandoahHeap::heap()->heuristics()->can_do_traversal_gc()) {

      Register thread = NOT_LP64(rax) LP64_ONLY(r15_thread);

#ifndef _LP64

      __ push(thread);

      __ get_thread(thread);

#endif

      Label done;

      // Short-circuit if count == 0.

      __ testptr(count, count);

      __ jcc(Assembler::zero, done);

      // Avoid runtime call when not marking.

      Address gc_state(thread, in_bytes(ShenandoahThreadLocalData::gc_state_offset()));

      __ testb(gc_state, ShenandoahHeap::MARKING);

      __ jcc(Assembler::zero, done);

      __ pusha();                      // push registers

#ifdef _LP64

      if (count == c_rarg0) {

        if (dst == c_rarg1) {

          // exactly backwards!!

          __ xchgptr(c_rarg1, c_rarg0);

        } else {

          __ movptr(c_rarg1, count);

          __ movptr(c_rarg0, dst);

        }

      } else {

        __ movptr(c_rarg0, dst);

        __ movptr(c_rarg1, count);

      }

      if (UseCompressedOops) {

        __ call_VM_leaf(CAST_FROM_FN_PTR(address, ShenandoahRuntime::write_ref_array_pre_narrow_oop_entry), 2);

      } else {

        __ call_VM_leaf(CAST_FROM_FN_PTR(address, ShenandoahRuntime::write_ref_array_pre_oop_entry), 2);

      }

#else

      __ call_VM_leaf(CAST_FROM_FN_PTR(address, ShenandoahRuntime::write_ref_array_pre_oop_entry),

                      dst, count);

#endif

      __ popa();

      __ bind(done);

      NOT_LP64(__ pop(thread);)

    }

  }

}

void ShenandoahBarrierSetAssembler::arraycopy_epilogue(MacroAssembler* masm, DecoratorSet decorators, BasicType type,

                                                       Register src, Register dst, Register count) {

  bool checkcast = (decorators & ARRAYCOPY_CHECKCAST) != 0;

  bool disjoint = (decorators & ARRAYCOPY_DISJOINT) != 0;

  bool obj_int = type == T_OBJECT LP64_ONLY(&& UseCompressedOops);

  Register tmp = rax;

  if (type == T_OBJECT || type == T_ARRAY) {

#ifdef _LP64

    if (!checkcast && !obj_int) {

      // Save count for barrier

      count = r11;

    } else if (disjoint && obj_int) {

      // Use the saved dst in the disjoint case

      dst = r11;

    } else if (checkcast) {

      tmp = rscratch1;

    }

#else

    if (disjoint) {

      __ mov(dst, rdx); // restore 'to'

    }

#endif

    __ pusha();             // push registers (overkill)

#ifdef _LP64

    if (c_rarg0 == count) { // On win64 c_rarg0 == rcx

      assert_different_registers(c_rarg1, dst);

      __ mov(c_rarg1, count);

      __ mov(c_rarg0, dst);

    } else {

      assert_different_registers(c_rarg0, count);

      __ mov(c_rarg0, dst);

      __ mov(c_rarg1, count);

    }

    __ call_VM_leaf(CAST_FROM_FN_PTR(address, ShenandoahRuntime::write_ref_array_post_entry), 2);

#else

    __ call_VM_leaf(CAST_FROM_FN_PTR(address, ShenandoahRuntime::write_ref_array_post_entry),

                    dst, count);

#endif

    __ popa();

  }

}

void ShenandoahBarrierSetAssembler::shenandoah_write_barrier_pre(MacroAssembler* masm,

                                                                 Register obj,

                                                                 Register pre_val,

                                                                 Register thread,

                                                                 Register tmp,

                                                                 bool tosca_live,

                                                                 bool expand_call) {

  if (ShenandoahSATBBarrier) {

    satb_write_barrier_pre(masm, obj, pre_val, thread, tmp, tosca_live, expand_call);

  }

}

void ShenandoahBarrierSetAssembler::satb_write_barrier_pre(MacroAssembler* masm,

                                                           Register 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.

#ifdef _LP64

  assert(thread == r15_thread, "must be");

#endif // _LP64

  Label done;

  Label runtime;

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

  if (obj != noreg) {

    assert_different_registers(obj, pre_val, tmp);

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

  }

  Address in_progress(thread, in_bytes(ShenandoahThreadLocalData::satb_mark_queue_active_offset()));

  Address index(thread, in_bytes(ShenandoahThreadLocalData::satb_mark_queue_index_offset()));

  Address buffer(thread, in_bytes(ShenandoahThreadLocalData::satb_mark_queue_buffer_offset()));

  Address gc_state(thread, in_bytes(ShenandoahThreadLocalData::gc_state_offset()));

  __ testb(gc_state, ShenandoahHeap::MARKING | ShenandoahHeap::TRAVERSAL);

  __ jcc(Assembler::zero, done);

  // Do we need to load the previous value?

  if (obj != noreg) {

    __ load_heap_oop(pre_val, Address(obj, 0), noreg, noreg, AS_RAW);

  }

  // Is the previous value null?

  __ cmpptr(pre_val, (int32_t) NULL_WORD);

  __ jcc(Assembler::equal, done);

  // Can we store original value in the thread's buffer?

  // Is index == 0?

  // (The index field is typed as size_t.)

  __ movptr(tmp, index);                   // tmp := *index_adr

  __ cmpptr(tmp, 0);                       // tmp == 0?

  __ jcc(Assembler::equal, runtime);       // If yes, goto runtime

  __ subptr(tmp, wordSize);                // tmp := tmp - wordSize

  __ movptr(index, tmp);                   // *index_adr := tmp

  __ addptr(tmp, buffer);                  // tmp := tmp + *buffer_adr

  // Record the previous value

  __ movptr(Address(tmp, 0), pre_val);

  __ jmp(done);

  __ bind(runtime);

  // save the live input values

  if(tosca_live) __ push(rax);

  if (obj != noreg && obj != rax)

    __ push(obj);

  if (pre_val != rax)

    __ push(pre_val);

  // Calling the runtime using the regular call_VM_leaf mechanism generates

  // code (generated by InterpreterMacroAssember::call_VM_leaf_base)

  // that checks that the *(ebp+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.

  NOT_LP64( __ push(thread); )

#ifdef _LP64

  // We move pre_val into c_rarg0 early, in order to avoid smashing it, should

  // pre_val be c_rarg1 (where the call prologue would copy thread argument).

  // Note: this should not accidentally smash thread, because thread is always r15.

  assert(thread != c_rarg0, "smashed arg");

  if (c_rarg0 != pre_val) {

    __ mov(c_rarg0, pre_val);

  }

#endif

  if (expand_call) {

    LP64_ONLY( assert(pre_val != c_rarg1, "smashed arg"); )

#ifdef _LP64

    if (c_rarg1 != thread) {

      __ mov(c_rarg1, thread);

    }

    // Already moved pre_val into c_rarg0 above

#else

    __ push(thread);

    __ push(pre_val);

#endif

    __ MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, ShenandoahRuntime::write_ref_field_pre_entry), 2);

  } else {

    __ call_VM_leaf(CAST_FROM_FN_PTR(address, ShenandoahRuntime::write_ref_field_pre_entry), LP64_ONLY(c_rarg0) NOT_LP64(pre_val), thread);

  }

  NOT_LP64( __ pop(thread); )

  // save the live input values

  if (pre_val != rax)

    __ pop(pre_val);

  if (obj != noreg && obj != rax)

    __ pop(obj);

  if(tosca_live) __ pop(rax);

  __ bind(done);

}

  Label is_null;

  __ testptr(dst, dst);

  __ jcc(Assembler::zero, is_null);

  __ bind(is_null);

}

  __ movptr(dst, Address(dst, ShenandoahBrooksPointer::byte_offset()));

}

#ifdef _LP64

  Label done;

  Address gc_state(r15_thread, in_bytes(ShenandoahThreadLocalData::gc_state_offset()));

  __ testb(gc_state, ShenandoahHeap::HAS_FORWARDED | ShenandoahHeap::EVACUATION | ShenandoahHeap::TRAVERSAL);

  __ jccb(Assembler::zero, done);

  __ testb(gc_state, ShenandoahHeap::EVACUATION | ShenandoahHeap::TRAVERSAL);

  __ jccb(Assembler::zero, done);

   if (dst != rax) {

     __ xchgptr(dst, rax); // Move obj into rax and save rax into obj.

   }

   if (dst != rax) {

     __ xchgptr(rax, dst); // Swap back obj with rax.

   }

  __ bind(done);

#else

  Unimplemented();

#endif

}

void ShenandoahBarrierSetAssembler::storeval_barrier(MacroAssembler* masm, Register dst, Register tmp) {

    storeval_barrier_impl(masm, dst, tmp);

  }

}

void ShenandoahBarrierSetAssembler::storeval_barrier_impl(MacroAssembler* masm, Register dst, Register tmp) {

  if (dst == noreg) return;

#ifdef _LP64

  if (ShenandoahStoreValEnqueueBarrier) {

    // The set of registers to be saved+restored is the same as in the write-barrier above.

    // Those are the commonly used registers in the interpreter.

    __ pusha();

    // __ push_callee_saved_registers();

    __ subptr(rsp, 2 * Interpreter::stackElementSize);

    __ movdbl(Address(rsp, 0), xmm0);

    satb_write_barrier_pre(masm, noreg, dst, r15_thread, tmp, true, false);

    __ movdbl(xmm0, Address(rsp, 0));

    __ addptr(rsp, 2 * Interpreter::stackElementSize);

    //__ pop_callee_saved_registers();

    __ popa();

  }

#else

  Unimplemented();

#endif

}

void ShenandoahBarrierSetAssembler::load_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;

  }

}

void ShenandoahBarrierSetAssembler::store_at(MacroAssembler* masm, DecoratorSet decorators, BasicType type,

              Address dst, Register val, Register tmp1, Register tmp2) {

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

  bool as_normal = (decorators & AS_NORMAL) != 0;

    bool needs_pre_barrier = as_normal;

    Register tmp3 = LP64_ONLY(r8) NOT_LP64(rsi);

    Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx);

    // flatten object address if needed

    // We do it regardless of precise because we need the registers

    if (dst.index() == noreg && dst.disp() == 0) {

      if (dst.base() != tmp1) {

        __ movptr(tmp1, dst.base());

      }

    } else {

      __ lea(tmp1, dst);

    }

#ifndef _LP64

    InterpreterMacroAssembler *imasm = static_cast<InterpreterMacroAssembler*>(masm);

#endif

    NOT_LP64(__ get_thread(rcx));

    NOT_LP64(imasm->save_bcp());

    if (needs_pre_barrier) {

      shenandoah_write_barrier_pre(masm /*masm*/,

                                   tmp1 /* obj */,

                                   tmp2 /* pre_val */,

                                   rthread /* thread */,

                                   tmp3  /* tmp */,

                                   val != noreg /* tosca_live */,

                                   false /* expand_call */);

    }

    if (val == noreg) {

      BarrierSetAssembler::store_at(masm, decorators, type, Address(tmp1, 0), val, noreg, noreg);

    } else {

      storeval_barrier(masm, val, tmp3);

      BarrierSetAssembler::store_at(masm, decorators, type, Address(tmp1, 0), val, noreg, noreg);

    }

    NOT_LP64(imasm->restore_bcp());

  } else {

    BarrierSetAssembler::store_at(masm, decorators, type, dst, val, tmp1, tmp2);

  }

}

void ShenandoahBarrierSetAssembler::tlab_allocate(MacroAssembler* masm,

                                                  Register thread, Register obj,

                                                  Register var_size_in_bytes,

                                                  int con_size_in_bytes,

                                                  Register t1, Register t2,

                                                  Label& slow_case) {

  assert_different_registers(obj, t1, t2);

  assert_different_registers(obj, var_size_in_bytes, t1);

  Register end = t2;

  if (!thread->is_valid()) {

#ifdef _LP64

    thread = r15_thread;

#else

    assert(t1->is_valid(), "need temp reg");

    thread = t1;

    __ get_thread(thread);

#endif

  }

  __ verify_tlab();

  __ movptr(obj, Address(thread, JavaThread::tlab_top_offset()));

  if (var_size_in_bytes == noreg) {

    __ lea(end, Address(obj, con_size_in_bytes + ShenandoahBrooksPointer::byte_size()));

  } else {

    __ addptr(var_size_in_bytes, ShenandoahBrooksPointer::byte_size());

    __ lea(end, Address(obj, var_size_in_bytes, Address::times_1));

  }

  __ cmpptr(end, Address(thread, JavaThread::tlab_end_offset()));

  __ jcc(Assembler::above, slow_case);

  // update the tlab top pointer

  __ movptr(Address(thread, JavaThread::tlab_top_offset()), end);

  // Initialize brooks pointer

#ifdef _LP64

  __ incrementq(obj, ShenandoahBrooksPointer::byte_size());

#else

  __ incrementl(obj, ShenandoahBrooksPointer::byte_size());

#endif

  __ movptr(Address(obj, ShenandoahBrooksPointer::byte_offset()), obj);

  // recover var_size_in_bytes if necessary

  if (var_size_in_bytes == end) {

    __ subptr(var_size_in_bytes, obj);

  }

  __ verify_tlab();

}

// Special Shenandoah CAS implementation that handles false negatives

// due to concurrent evacuation.

#ifndef _LP64

void ShenandoahBarrierSetAssembler::cmpxchg_oop(MacroAssembler* masm,

                                                Register res, Address addr, Register oldval, Register newval,

                                                bool exchange, Register tmp1, Register tmp2) {

  // Shenandoah has no 32-bit version for this.

  Unimplemented();

}

#else

void ShenandoahBarrierSetAssembler::cmpxchg_oop(MacroAssembler* masm,

                                                Register res, Address addr, Register oldval, Register newval,

                                                bool exchange, Register tmp1, Register tmp2) {

  assert(ShenandoahCASBarrier, "Should only be used when CAS barrier is enabled");

  assert(oldval == rax, "must be in rax for implicit use in cmpxchg");