/*

 * 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/shenandoah/shenandoahBarrierSetAssembler.hpp"

#include "gc/shenandoah/shenandoahForwarding.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->

address ShenandoahBarrierSetAssembler::_shenandoah_lrb = NULL;

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) {

      if (!obj_int) {

        // Save count for barrier

        __ movptr(r11, count);

      } else if (disjoint) {

        // Save dst in r11 in the disjoint case

        __ movq(r11, dst);

      }

    }

#else

    if (disjoint) {

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

    }

#endif

    if (ShenandoahSATBBarrier && !dest_uninitialized) {

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

      assert_different_registers(dst, count, thread); // we don't care about src here?

#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) {

      if (!obj_int) {

        // Save count for barrier

        count = r11;

      } else if (disjoint && obj_int) {

        // Use the saved dst in the disjoint case

        dst = r11;

      }

    } else {

      tmp = rscratch1;

    }

#else

    if (disjoint) {

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

    }

#endif

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

    assert_different_registers(dst, thread); // do we care about src at all here?

#ifndef _LP64

    __ push(thread);

    __ get_thread(thread);

#endif

    // Short-circuit if count == 0.

    Label done;

    __ testptr(count, count);

    __ jcc(Assembler::zero, done);

    // Skip runtime call if no forwarded objects.

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

    __ testb(gc_state, ShenandoahHeap::UPDATEREFS);

    __ jcc(Assembler::zero, done);

    __ 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();

    __ bind(done);

    NOT_LP64(__ pop(thread);)

  }

}

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);

}

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

  assert(ShenandoahCASBarrier, "should be enabled");

  Label is_null;

  __ testptr(dst, dst);

  __ jcc(Assembler::zero, is_null);

  resolve_forward_pointer_not_null(masm, dst, tmp);

  __ bind(is_null);

}

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

  assert(ShenandoahCASBarrier || ShenandoahLoadRefBarrier, "should be enabled");

  // The below loads the mark word, checks if the lowest two bits are

  // set, and if so, clear the lowest two bits and copy the result

  // to dst. Otherwise it leaves dst alone.

  // Implementing this is surprisingly awkward. I do it here by:

  // - Inverting the mark word

  // - Test lowest two bits == 0

  // - If so, set the lowest two bits

  // - Invert the result back, and copy to dst

  bool borrow_reg = (tmp == noreg);

  if (borrow_reg) {

    // No free registers available. Make one useful.

    tmp = LP64_ONLY(rscratch1) NOT_LP64(rdx);

    __ push(tmp);

  }

  Label done;

  __ movptr(tmp, Address(dst, oopDesc::mark_offset_in_bytes()));

  __ notptr(tmp);

  __ testb(tmp, markOopDesc::marked_value);

  __ jccb(Assembler::notZero, done);

  __ orptr(tmp, markOopDesc::marked_value);

  __ notptr(tmp);

  __ mov(dst, tmp);

  __ bind(done);

  if (borrow_reg) {

    __ pop(tmp);

  }

}

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

  assert(ShenandoahLoadRefBarrier, "Should be enabled");

  Label done;

#ifdef _LP64

  Register thread = r15_thread;

#else

  Register thread = rcx;

  if (thread == dst) {

    thread = rbx;

  }

  __ push(thread);

  __ get_thread(thread);

#endif

  assert_different_registers(dst, thread);

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

  __ testb(gc_state, ShenandoahHeap::HAS_FORWARDED);

  __ jccb(Assembler::zero, done);

   if (dst != rax) {

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

   }

   __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, ShenandoahBarrierSetAssembler::shenandoah_lrb())));

   if (dst != rax) {

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

   }

  __ bind(done);

#ifndef _LP64

  __ pop(thread);

#endif

}

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

  if (ShenandoahStoreValEnqueueBarrier) {

    storeval_barrier_impl(masm, dst, tmp);

  }

}

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

  assert(ShenandoahStoreValEnqueueBarrier, "should be enabled");

  if (dst == noreg) return;

  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);

#ifdef _LP64

    Register thread = r15_thread;

#else

    Register thread = rcx;

    if (thread == dst || thread == tmp) {

      thread = rdi;

    }

    if (thread == dst || thread == tmp) {

      thread = rbx;