/*
* This file is part of the Jikes RVM project (http://jikesrvm.org).
*
* This file is licensed to You under the Common Public License (CPL);
* You may not use this file except in compliance with the License. You
* may obtain a copy of the License at
*
* http://www.opensource.org/licenses/cpl1.0.php
*
* See the COPYRIGHT.txt file distributed with this work for information
* regarding copyright ownership.
*/
package org.mmtk.plan;
import org.jikesrvm.annotations.NoSubArchCompile;
import org.mmtk.policy.MarkSweepLocal;
import org.mmtk.policy.Space;
import org.mmtk.policy.ImmortalLocal;
import org.mmtk.policy.LargeObjectLocal;
import org.mmtk.utility.alloc.Allocator;
import org.mmtk.utility.alloc.BumpPointer;
import org.mmtk.utility.Constants;
import org.mmtk.utility.Log;
import org.mmtk.vm.VM;
import org.vmmagic.pragma.*;
import org.vmmagic.unboxed.*;
/**
* This class (and its sub-classes) implement <i>per-mutator thread</i>
* behavior. We assume <i>N</i> collector threads and <i>M</i>
* mutator threads, where <i>N</i> is often equal to the number of
* available processors, P (for P-way parallelism at GC-time), and
* <i>M</i> may simply be the number of mutator (application) threads.
* Both <i>N</i> and <i>M</i> are determined by the VM, not MMTk. In
* the case where a VM uses posix threads (pthreads) for each mutator
* ("1:1" threading), <i>M</i> will typically be equal to the number of
* mutator threads. When a uses "green threads" or a hybrid threading
* scheme (such as Jikes RVM), <i>M</i> will typically be equal to the
* level of <i>true</i> parallelism (ie the number of underlying
* kernel threads).<p>
*
* MMTk assumes that the VM instantiates instances of MutatorContext
* in thread local storage (TLS) for each thread participating in
* collection. Accesses to this state are therefore assumed to be
* low-cost during mutator time.<p>
*
* This class (and its children) is therefore used for unsynchronized
* per-mutator operations such as <i>allocation</i> and <i>write barriers</i>.
* The semantics and necessary state for these operations are therefore
* specified in the GC-specific subclasses of this class.
*
* MMTk explicitly separates thread-local (this class) and global
* operations (@see Plan), so that syncrhonization is localized
* and explicit, and thus hopefully minimized (@see Plan). Gloabl (Plan)
* and per-thread (this class) state are also explicitly separated.
* Operations in this class (and its children) are therefore strictly
* local to each mutator thread, and synchronized operations always
* happen via access to explicitly global classes such as Plan and its
* children. Therefore only <i>"fast path"</i> (unsynchronized)
* allocation and barrier semantics are defined in MutatorContext and
* its subclasses. These call out to <i>"slow path"</i> (synchronize(d)
* methods which have global state and are globally synchronized. For
* example, an allocation fast path may bump a pointer without any
* syncrhonization (the "fast path") until a limit is reached, at which
* point the "slow path" is called, and more memory is aquired from a
* global resource.<p>
*
* As the super-class of all per-mutator contexts, this class implements
* basic per-mutator behavior common to all MMTk collectors, including
* support for immortal and large object space allocation, as well as
* empty stubs for write barriers (to be overridden by sub-classes as
* needed).
*
* @see CollectorContext
* @see org.mmtk.vm.ActivePlan
* @see Plan
*/
@Uninterruptible
public abstract class MutatorContext implements Constants {
/****************************************************************************
* Instance fields
*/
/** Unique mutator identifier */
protected int id = VM.activePlan.registerMutator(this);
/** Used for printing log information in a thread safe manner */
protected final Log log = new Log();
/** Per-mutator allocator into the immortal space */
protected BumpPointer immortal = new ImmortalLocal(Plan.immortalSpace);
/** Per-mutator allocator into the large object space */
protected LargeObjectLocal los = new LargeObjectLocal(Plan.loSpace);
/** Per-mutator allocator into the small code space */
private MarkSweepLocal smcode = Plan.USE_CODE_SPACE ? new MarkSweepLocal(Plan.smallCodeSpace) : null;
/** Per-mutator allocator into the large code space */
private LargeObjectLocal lgcode = Plan.USE_CODE_SPACE ? new LargeObjectLocal(Plan.largeCodeSpace) : null;
/** Per-mutator allocator into the primitive large object space */
protected LargeObjectLocal plos = new LargeObjectLocal(Plan.ploSpace);
/****************************************************************************
*
* Collection.
*/
/**
* Perform a per-mutator collection phase.
*
* @param phaseId The unique phase identifier
* @param primary Should this thread be used to execute any single-threaded
* local operations?
*/
public abstract void collectionPhase(short phaseId, boolean primary);
/****************************************************************************
*
* Allocation.
*/
/**
* Run-time check of the allocator to use for a given allocation
*
* At the moment this method assumes that allocators will use the simple
* (worst) method of aligning to determine if the object is a large object
* to ensure that no objects are larger than other allocators can handle.
*
* @param bytes The number of bytes to be allocated
* @param align The requested alignment.
* @param allocator The allocator statically assigned to this allocation
* @return The allocator dyncamically assigned to this allocation
*/
@Inline
public int checkAllocator(int bytes, int align, int allocator) {
if (allocator == Plan.ALLOC_DEFAULT &&
Allocator.getMaximumAlignedSize(bytes, align) > Plan.LOS_SIZE_THRESHOLD)
return Plan.ALLOC_LOS;
else if (Plan.USE_CODE_SPACE && allocator == Plan.ALLOC_CODE) {
if (Allocator.getMaximumAlignedSize(bytes, align) > Plan.LOS_SIZE_THRESHOLD)
return Plan.ALLOC_LARGE_CODE;
else
return allocator;
} else if (allocator == Plan.ALLOC_NON_REFERENCE) {
if (Allocator.getMaximumAlignedSize(bytes, align) > Plan.PLOS_SIZE_THRESHOLD)
return Plan.ALLOC_PRIMITIVE_LOS;
else
return Plan.ALLOC_DEFAULT;
} else
return allocator;
}
/**
* Allocate memory for an object.
*
* @param bytes The number of bytes required for the object.
* @param align Required alignment for the object.
* @param offset Offset associated with the alignment.
* @param allocator The allocator associated with this request.
* @param site Allocation site
* @return The low address of the allocated chunk.
*/
@Inline
public Address alloc(int bytes, int align, int offset, int allocator, int site) {
switch (allocator) {
case Plan.ALLOC_LOS: return los.alloc(bytes, align, offset);
case Plan.ALLOC_PRIMITIVE_LOS: return plos.alloc(bytes, align, offset);
case Plan.ALLOC_IMMORTAL: return immortal.alloc(bytes, align, offset);
case Plan.ALLOC_CODE: return smcode.alloc(bytes, align, offset);
case Plan.ALLOC_LARGE_CODE: return lgcode.alloc(bytes, align, offset);
default:
VM.assertions.fail("No such allocator");
return Address.zero();
}
}
/**
* Perform post-allocation actions. For many allocators none are
* required.
*
* @param ref The newly allocated object
* @param typeRef the type reference for the instance being created
* @param bytes The size of the space to be allocated (in bytes)
* @param allocator The allocator number to be used for this allocation
*/
@Inline
public void postAlloc(ObjectReference ref, ObjectReference typeRef,
int bytes, int allocator) {
switch (allocator) {
case Plan.ALLOC_LOS: Plan.loSpace.initializeHeader(ref, false); return;
case Plan.ALLOC_PRIMITIVE_LOS: Plan.ploSpace.initializeHeader(ref, true); return;
case Plan.ALLOC_IMMORTAL: Plan.immortalSpace.initializeHeader(ref); return;
case Plan.ALLOC_CODE: Plan.smallCodeSpace.initializeHeader(ref, true); return;
case Plan.ALLOC_LARGE_CODE: Plan.largeCodeSpace.initializeHeader(ref, true); return;
default:
VM.assertions.fail("No such allocator");
}
}
/****************************************************************************
*
* Space - Allocator mapping. See description for getOwnAllocator that
* describes why this is important.
*/
/**
* Given an allocator, <code>a</code>, determine the space into
* which <code>a</code> is allocating and then return an allocator
* (possibly <code>a</code>) associated with <i>this plan
* instance</i> which is allocating into the same space as
* <code>a</code>.<p>
*
* The need for the method is subtle. The problem arises because
* application threads may change their affinity with
* processors/posix threads, and this may happen during a GC (at the
* point at which the scheduler performs thread switching associated
* with the GC). At the end of a GC, the thread that triggered the
* GC may now be bound to a different processor and thus the
* allocator instance on its stack may be no longer be valid
* (i.e. it may pertain to a different plan instance).<p>
*
* This method allows the correct allocator instance to be
* established and associated with the thread (see {@link
* org.mmtk.utility.alloc.Allocator#allocSlow(int, int, int) Allocator.allocSlow()}).
*
* @see org.mmtk.utility.alloc.Allocator
* @see org.mmtk.utility.alloc.Allocator#allocSlow(int, int, int)
*
* @param a An allocator instance.
* @return An allocator instance associated with <i>this plan
* instance</i> that allocates into the same space as <code>a</code>
* (this may in fact be <code>a</code>).
*/
public final Allocator getOwnAllocator(Allocator a) {
Space space = Plan.getSpaceFromAllocatorAnyLocal(a);
if (space == null)
VM.assertions.fail("MutatorContext.getOwnAllocator could not obtain space");
return getAllocatorFromSpace(space);
}
/**
* Return the space into which an allocator is allocating. This
* particular method will match against those spaces defined at this
* level of the class hierarchy. Subclasses must deal with spaces
* they define and refer to superclasses appropriately.
*
* @param a An allocator
* @return The space into which <code>a</code> is allocating, or
* <code>null</code> if there is no space associated with
* <code>a</code>.
*/
public Space getSpaceFromAllocator(Allocator a) {
if (a == immortal) return Plan.immortalSpace;
if (a == los) return Plan.loSpace;
if (a == plos) return Plan.ploSpace;
if (Plan.USE_CODE_SPACE && a == smcode) return Plan.smallCodeSpace;
if (Plan.USE_CODE_SPACE && a == lgcode) return Plan.largeCodeSpace;
// a does not belong to this plan instance
return null;
}
/**
* Return the allocator instance associated with a space
* <code>space</code>, for this plan instance.
*
* @param space The space for which the allocator instance is desired.
* @return The allocator instance associated with this plan instance
* which is allocating into <code>space</code>, or <code>null</code>
* if no appropriate allocator can be established.
*/
public Allocator getAllocatorFromSpace(Space space) {
if (space == Plan.immortalSpace) return immortal;
if (space == Plan.loSpace) return los;
if (space == Plan.ploSpace) return plos;
if (Plan.USE_CODE_SPACE && space == Plan.smallCodeSpace) return smcode;
if (Plan.USE_CODE_SPACE && space == Plan.largeCodeSpace) return lgcode;
// Invalid request has been made
if (space == Plan.metaDataSpace) {
VM.assertions.fail("MutatorContext.getAllocatorFromSpace given meta space");
} else if (space != null) {
VM.assertions.fail("MutatorContext.getAllocatorFromSpace given invalid space");
} else {
VM.assertions.fail("MutatorContext.getAllocatorFromSpace given null space");
}
return null;
}
/****************************************************************************
*
* Write and read barriers. By default do nothing, override if
* appropriate.
*/
/**
* A new reference is about to be created. Take appropriate write
* barrier actions.<p>
*
* <b>By default do nothing, override if appropriate.</b>
*
* @param src The object into which the new reference will be stored
* @param slot The address into which the new reference will be
* stored.
* @param tgt The target of the new reference
* @param metaDataA An int that assists the host VM in creating a store
* @param metaDataB An int that assists the host VM in creating a store
* @param mode The context in which the store occured
*/
public void writeBarrier(ObjectReference src, Address slot,
ObjectReference tgt, Offset metaDataA,
int metaDataB, int mode) {
// Either: write barriers are used and this is overridden, or
// write barriers are not used and this is never called
if (VM.VERIFY_ASSERTIONS) VM.assertions._assert(false);
}
/**
* Attempt to atomically exchange the value in the given slot
* with the passed replacement value. If a new reference is
* created, we must then take appropriate write barrier actions.<p>
*
* <b>By default do nothing, override if appropriate.</b>
*
* @param src The object into which the new reference will be stored
* @param slot The address into which the new reference will be
* stored.
* @param old The old reference to be swapped out
* @param tgt The target of the new reference
* @param metaDataA An int that assists the host VM in creating a store
* @param metaDataB An int that assists the host VM in creating a store
* @param mode The context in which the store occured
* @return True if the swap was successful.
*/
public boolean tryCompareAndSwapWriteBarrier(ObjectReference src, Address slot,
ObjectReference old, ObjectReference tgt, Offset metaDataA,
int metaDataB, int mode) {
// Either: write barriers are used and this is overridden, or
// write barriers are not used and this is never called
if (VM.VERIFY_ASSERTIONS) VM.assertions._assert(false);
return false;
}
/**
* A number of references are about to be copied from object
* <code>src</code> to object <code>dst</code> (as in an array
* copy). Thus, <code>dst</code> is the mutated object. Take
* appropriate write barrier actions.<p>
*
* @param src The source of the values to be copied
* @param srcOffset The offset of the first source address, in
* bytes, relative to <code>src</code> (in principle, this could be
* negative).
* @param dst The mutated object, i.e. the destination of the copy.
* @param dstOffset The offset of the first destination address, in
* bytes relative to <code>tgt</code> (in principle, this could be
* negative).
* @param bytes The size of the region being copied, in bytes.
* @return True if the update was performed by the barrier, false if
* left to the caller (always false in this case).
*/
public boolean writeBarrier(ObjectReference src, Offset srcOffset,
ObjectReference dst, Offset dstOffset,
int bytes) {
// Either: write barriers are used and this is overridden, or
// write barriers are not used and this is never called
if (VM.VERIFY_ASSERTIONS) VM.assertions._assert(false);
return false;
}
/**
* Read a reference type. In a concurrent collector this may
* involve adding the referent to the marking queue.
*
* @param referent The referent being read.
* @return The new referent.
*/
@Inline
public ObjectReference referenceTypeReadBarrier(ObjectReference referent) {
if (VM.VERIFY_ASSERTIONS) VM.assertions._assert(false);
return ObjectReference.nullReference();
}
/**
* Read a reference. Take appropriate read barrier action, and
* return the value that was read.<p> This is a <b>substituting<b>
* barrier. The call to this barrier takes the place of a load.<p>
*
* @param src The object reference being read.
* @param context The context in which the read arose (getfield, for example)
* @return The reference that was read.
*/
@Inline
public Address readBarrier(ObjectReference src, Address slot,
int context) {
// read barrier currently unimplemented
if (VM.VERIFY_ASSERTIONS) VM.assertions._assert(false);
return Address.max();
}
/**
* Flush mutator context, in response to a requestMutatorFlush
*/
public void flush() {
flushRememberedSets();
}
/**
* Flush per-mutator remembered sets into the global remset pool.
*/
public void flushRememberedSets() {
// Either: write barriers are used and this is overridden, or
// write barriers are not used and this is a no-op
}
/**
* Assert that the remsets have been flushed. This is critical to
* correctness. We need to maintain the invariant that remset entries
* do not accrue during GC. If the host JVM generates barrier entires
* it is its own responsibility to ensure that they are flushed before
* returning to MMTk.
*/
public void assertRemsetsFlushed() {
// Either: write barriers are used and this is overridden, or
// write barriers are not used and this is a no-op
}
/***********************************************************************
*
* Miscellaneous
*/
/** @return the unique identifier for this mutator context. */
@Inline
public int getId() { return id; }
}