/*
- * Copyright (c) 2012-2014 ARM Limited
+ * Copyright (c) 2012-2015 ARM Limited
* All rights reserved
*
* The license below extends only to copyright in the software and shall
* modified or unmodified, in source code or in binary form.
*
* Copyright (c) 2006 The Regents of The University of Michigan
- * Copyright (c) 2010 Advanced Micro Devices, Inc.
+ * Copyright (c) 2010,2015 Advanced Micro Devices, Inc.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
ReadRespWithInvalidate,
WriteReq,
WriteResp,
- Writeback,
+ WritebackDirty,
+ WritebackClean,
+ CleanEvict,
SoftPFReq,
HardPFReq,
SoftPFResp,
HardPFResp,
- WriteInvalidateReq,
- WriteInvalidateResp,
+ WriteLineReq,
UpgradeReq,
SCUpgradeReq, // Special "weak" upgrade for StoreCond
UpgradeResp,
UpgradeFailResp, // Valid for SCUpgradeReq only
ReadExReq,
ReadExResp,
+ ReadCleanReq,
+ ReadSharedReq,
LoadLockedReq,
StoreCondReq,
StoreCondFailReq, // Failed StoreCondReq in MSHR (never sent)
SwapResp,
MessageReq,
MessageResp,
+ MemFenceReq,
+ MemFenceResp,
// Error responses
// @TODO these should be classified as responses rather than
// requests; coding them as requests initially for backwards
// Fake simulator-only commands
PrintReq, // Print state matching address
FlushReq, //request for a cache flush
- InvalidationReq, // request for address to be invalidated from lsq
+ InvalidateReq, // request for address to be invalidated
+ InvalidateResp,
NUM_MEM_CMDS
};
IsWrite, //!< Data flows from requester to responder
IsUpgrade,
IsInvalidate,
- NeedsExclusive, //!< Requires exclusive copy to complete in-cache
+ NeedsWritable, //!< Requires writable copy to complete in-cache
IsRequest, //!< Issued by requester
IsResponse, //!< Issue by responder
NeedsResponse, //!< Requester needs response from target
+ IsEviction,
IsSWPrefetch,
IsHWPrefetch,
IsLlsc, //!< Alpha/MIPS LL or SC access
public:
- bool isRead() const { return testCmdAttrib(IsRead); }
- bool isWrite() const { return testCmdAttrib(IsWrite); }
- bool isUpgrade() const { return testCmdAttrib(IsUpgrade); }
- bool isRequest() const { return testCmdAttrib(IsRequest); }
- bool isResponse() const { return testCmdAttrib(IsResponse); }
- bool needsExclusive() const { return testCmdAttrib(NeedsExclusive); }
- bool needsResponse() const { return testCmdAttrib(NeedsResponse); }
- bool isInvalidate() const { return testCmdAttrib(IsInvalidate); }
+ bool isRead() const { return testCmdAttrib(IsRead); }
+ bool isWrite() const { return testCmdAttrib(IsWrite); }
+ bool isUpgrade() const { return testCmdAttrib(IsUpgrade); }
+ bool isRequest() const { return testCmdAttrib(IsRequest); }
+ bool isResponse() const { return testCmdAttrib(IsResponse); }
+ bool needsWritable() const { return testCmdAttrib(NeedsWritable); }
+ bool needsResponse() const { return testCmdAttrib(NeedsResponse); }
+ bool isInvalidate() const { return testCmdAttrib(IsInvalidate); }
+ bool isEviction() const { return testCmdAttrib(IsEviction); }
+
+ /**
+ * A writeback is an eviction that carries data.
+ */
+ bool isWriteback() const { return testCmdAttrib(IsEviction) &&
+ testCmdAttrib(HasData); }
+
+ /**
+ * Check if this particular packet type carries payload data. Note
+ * that this does not reflect if the data pointer of the packet is
+ * valid or not.
+ */
bool hasData() const { return testCmdAttrib(HasData); }
bool isLLSC() const { return testCmdAttrib(IsLlsc); }
bool isSWPrefetch() const { return testCmdAttrib(IsSWPrefetch); }
bool isPrint() const { return testCmdAttrib(IsPrint); }
bool isFlush() const { return testCmdAttrib(IsFlush); }
- const Command
+ Command
responseCommand() const
{
return commandInfo[cmd].response;
typedef ::Flags<FlagsType> Flags;
private:
- static const FlagsType PUBLIC_FLAGS = 0x00000000;
- static const FlagsType PRIVATE_FLAGS = 0x00007F0F;
- static const FlagsType COPY_FLAGS = 0x0000000F;
-
- static const FlagsType SHARED = 0x00000001;
- // Special control flags
- /// Special timing-mode atomic snoop for multi-level coherence.
- static const FlagsType EXPRESS_SNOOP = 0x00000002;
- /// Does supplier have exclusive copy?
- /// Useful for multi-level coherence.
- static const FlagsType SUPPLY_EXCLUSIVE = 0x00000004;
- // Snoop response flags
- static const FlagsType MEM_INHIBIT = 0x00000008;
- /// Are the 'addr' and 'size' fields valid?
- static const FlagsType VALID_ADDR = 0x00000100;
- static const FlagsType VALID_SIZE = 0x00000200;
- /// Is the data pointer set to a value that shouldn't be freed
- /// when the packet is destroyed?
- static const FlagsType STATIC_DATA = 0x00001000;
- /// The data pointer points to a value that should be freed when
- /// the packet is destroyed.
- static const FlagsType DYNAMIC_DATA = 0x00002000;
- /// the data pointer points to an array (thus delete []) needs to
- /// be called on it rather than simply delete.
- static const FlagsType ARRAY_DATA = 0x00004000;
- /// suppress the error if this packet encounters a functional
- /// access failure.
- static const FlagsType SUPPRESS_FUNC_ERROR = 0x00008000;
- // Signal prefetch squash through express snoop flag
- static const FlagsType PREFETCH_SNOOP_SQUASH = 0x00010000;
+
+ enum : FlagsType {
+ // Flags to transfer across when copying a packet
+ COPY_FLAGS = 0x0000000F,
+
+ // Does this packet have sharers (which means it should not be
+ // considered writable) or not. See setHasSharers below.
+ HAS_SHARERS = 0x00000001,
+
+ // Special control flags
+ /// Special timing-mode atomic snoop for multi-level coherence.
+ EXPRESS_SNOOP = 0x00000002,
+
+ /// Allow a responding cache to inform the cache hierarchy
+ /// that it had a writable copy before responding. See
+ /// setResponderHadWritable below.
+ RESPONDER_HAD_WRITABLE = 0x00000004,
+
+ // Snoop co-ordination flag to indicate that a cache is
+ // responding to a snoop. See setCacheResponding below.
+ CACHE_RESPONDING = 0x00000008,
+
+ /// Are the 'addr' and 'size' fields valid?
+ VALID_ADDR = 0x00000100,
+ VALID_SIZE = 0x00000200,
+
+ /// Is the data pointer set to a value that shouldn't be freed
+ /// when the packet is destroyed?
+ STATIC_DATA = 0x00001000,
+ /// The data pointer points to a value that should be freed when
+ /// the packet is destroyed. The pointer is assumed to be pointing
+ /// to an array, and delete [] is consequently called
+ DYNAMIC_DATA = 0x00002000,
+
+ /// suppress the error if this packet encounters a functional
+ /// access failure.
+ SUPPRESS_FUNC_ERROR = 0x00008000,
+
+ // Signal block present to squash prefetch and cache evict packets
+ // through express snoop flag
+ BLOCK_CACHED = 0x00010000
+ };
Flags flags;
MemCmd cmd;
/// A pointer to the original request.
- RequestPtr req;
+ const RequestPtr req;
private:
/**
unsigned size;
/**
- * Source port identifier set on a request packet to enable
- * appropriate routing of the responses. The source port
- * identifier is set by any multiplexing component, e.g. a
- * crossbar, as the timing responses need this information to be
- * routed back to the appropriate port at a later point in
- * time. The field can be updated (over-written) as the request
- * packet passes through additional multiplexing components, and
- * it is their responsibility to remember the original source port
- * identifier, for example by using an appropriate sender
- * state. The latter is done in the cache and bridge.
- */
- PortID src;
-
- /**
- * Destination port identifier that is present on all response
- * packets that passed through a multiplexing component as a
- * request packet. The source port identifier is turned into a
- * destination port identifier when the packet is turned into a
- * response, and the destination is used, e.g. by the crossbar, to
- * select the appropriate path through the interconnect.
- */
- PortID dest;
-
- /**
- * The original value of the command field. Only valid when the
- * current command field is an error condition; in that case, the
- * previous contents of the command field are copied here. This
- * field is *not* set on non-error responses.
+ * Track the bytes found that satisfy a functional read.
*/
- MemCmd origCmd;
-
- /**
- * These values specify the range of bytes found that satisfy a
- * functional read.
- */
- uint16_t bytesValidStart;
- uint16_t bytesValidEnd;
+ std::vector<bool> bytesValid;
public:
/**
- * The extra delay from seeing the packet until the first word is
+ * The extra delay from seeing the packet until the header is
* transmitted. This delay is used to communicate the crossbar
* forwarding latency to the neighbouring object (e.g. a cache)
* that actually makes the packet wait. As the delay is relative,
* a 32-bit unsigned should be sufficient.
*/
- uint32_t firstWordDelay;
+ uint32_t headerDelay;
/**
- * The extra pipelining delay from seeing the packet until the
- * last word is transmitted by the component that provided it (if
- * any). This includes the first word delay. Similar to the first
- * word delay, this is used to make up for the fact that the
+ * Keep track of the extra delay incurred by snooping upwards
+ * before sending a request down the memory system. This is used
+ * by the coherent crossbar to account for the additional request
+ * delay.
+ */
+ uint32_t snoopDelay;
+
+ /**
+ * The extra pipelining delay from seeing the packet until the end of
+ * payload is transmitted by the component that provided it (if
+ * any). This includes the header delay. Similar to the header
+ * delay, this is used to make up for the fact that the
* crossbar does not make the packet wait. As the delay is
* relative, a 32-bit unsigned should be sufficient.
*/
- uint32_t lastWordDelay;
+ uint32_t payloadDelay;
/**
* A virtual base opaque structure used to hold state associated
/// Return the index of this command.
inline int cmdToIndex() const { return cmd.toInt(); }
- bool isRead() const { return cmd.isRead(); }
- bool isWrite() const { return cmd.isWrite(); }
- bool isUpgrade() const { return cmd.isUpgrade(); }
- bool isRequest() const { return cmd.isRequest(); }
- bool isResponse() const { return cmd.isResponse(); }
- bool needsExclusive() const { return cmd.needsExclusive(); }
- bool needsResponse() const { return cmd.needsResponse(); }
- bool isInvalidate() const { return cmd.isInvalidate(); }
- bool hasData() const { return cmd.hasData(); }
- bool isLLSC() const { return cmd.isLLSC(); }
- bool isError() const { return cmd.isError(); }
- bool isPrint() const { return cmd.isPrint(); }
- bool isFlush() const { return cmd.isFlush(); }
-
- // Snoop flags
- void assertMemInhibit() { flags.set(MEM_INHIBIT); }
- bool memInhibitAsserted() const { return flags.isSet(MEM_INHIBIT); }
- void assertShared() { flags.set(SHARED); }
- bool sharedAsserted() const { return flags.isSet(SHARED); }
-
- // Special control flags
- void setExpressSnoop() { flags.set(EXPRESS_SNOOP); }
- bool isExpressSnoop() const { return flags.isSet(EXPRESS_SNOOP); }
- void setSupplyExclusive() { flags.set(SUPPLY_EXCLUSIVE); }
- void clearSupplyExclusive() { flags.clear(SUPPLY_EXCLUSIVE); }
- bool isSupplyExclusive() const { return flags.isSet(SUPPLY_EXCLUSIVE); }
+ bool isRead() const { return cmd.isRead(); }
+ bool isWrite() const { return cmd.isWrite(); }
+ bool isUpgrade() const { return cmd.isUpgrade(); }
+ bool isRequest() const { return cmd.isRequest(); }
+ bool isResponse() const { return cmd.isResponse(); }
+ bool needsWritable() const
+ {
+ // we should never check if a response needsWritable, the
+ // request has this flag, and for a response we should rather
+ // look at the hasSharers flag (if not set, the response is to
+ // be considered writable)
+ assert(isRequest());
+ return cmd.needsWritable();
+ }
+ bool needsResponse() const { return cmd.needsResponse(); }
+ bool isInvalidate() const { return cmd.isInvalidate(); }
+ bool isEviction() const { return cmd.isEviction(); }
+ bool isWriteback() const { return cmd.isWriteback(); }
+ bool hasData() const { return cmd.hasData(); }
+ bool hasRespData() const
+ {
+ MemCmd resp_cmd = cmd.responseCommand();
+ return resp_cmd.hasData();
+ }
+ bool isLLSC() const { return cmd.isLLSC(); }
+ bool isError() const { return cmd.isError(); }
+ bool isPrint() const { return cmd.isPrint(); }
+ bool isFlush() const { return cmd.isFlush(); }
+
+ //@{
+ /// Snoop flags
+ /**
+ * Set the cacheResponding flag. This is used by the caches to
+ * signal another cache that they are responding to a request. A
+ * cache will only respond to snoops if it has the line in either
+ * Modified or Owned state. Note that on snoop hits we always pass
+ * the line as Modified and never Owned. In the case of an Owned
+ * line we proceed to invalidate all other copies.
+ *
+ * On a cache fill (see Cache::handleFill), we check hasSharers
+ * first, ignoring the cacheResponding flag if hasSharers is set.
+ * A line is consequently allocated as:
+ *
+ * hasSharers cacheResponding state
+ * true false Shared
+ * true true Shared
+ * false false Exclusive
+ * false true Modified
+ */
+ void setCacheResponding()
+ {
+ assert(isRequest());
+ assert(!flags.isSet(CACHE_RESPONDING));
+ flags.set(CACHE_RESPONDING);
+ }
+ bool cacheResponding() const { return flags.isSet(CACHE_RESPONDING); }
+ /**
+ * On fills, the hasSharers flag is used by the caches in
+ * combination with the cacheResponding flag, as clarified
+ * above. If the hasSharers flag is not set, the packet is passing
+ * writable. Thus, a response from a memory passes the line as
+ * writable by default.
+ *
+ * The hasSharers flag is also used by upstream caches to inform a
+ * downstream cache that they have the block (by calling
+ * setHasSharers on snoop request packets that hit in upstream
+ * cachs tags or MSHRs). If the snoop packet has sharers, a
+ * downstream cache is prevented from passing a dirty line upwards
+ * if it was not explicitly asked for a writable copy. See
+ * Cache::satisfyCpuSideRequest.
+ *
+ * The hasSharers flag is also used on writebacks, in
+ * combination with the WritbackClean or WritebackDirty commands,
+ * to allocate the block downstream either as:
+ *
+ * command hasSharers state
+ * WritebackDirty false Modified
+ * WritebackDirty true Owned
+ * WritebackClean false Exclusive
+ * WritebackClean true Shared
+ */
+ void setHasSharers() { flags.set(HAS_SHARERS); }
+ bool hasSharers() const { return flags.isSet(HAS_SHARERS); }
+ //@}
+
+ /**
+ * The express snoop flag is used for two purposes. Firstly, it is
+ * used to bypass flow control for normal (non-snoop) requests
+ * going downstream in the memory system. In cases where a cache
+ * is responding to a snoop from another cache (it had a dirty
+ * line), but the line is not writable (and there are possibly
+ * other copies), the express snoop flag is set by the downstream
+ * cache to invalidate all other copies in zero time. Secondly,
+ * the express snoop flag is also set to be able to distinguish
+ * snoop packets that came from a downstream cache, rather than
+ * snoop packets from neighbouring caches.
+ */
+ void setExpressSnoop() { flags.set(EXPRESS_SNOOP); }
+ bool isExpressSnoop() const { return flags.isSet(EXPRESS_SNOOP); }
+
+ /**
+ * On responding to a snoop request (which only happens for
+ * Modified or Owned lines), make sure that we can transform an
+ * Owned response to a Modified one. If this flag is not set, the
+ * responding cache had the line in the Owned state, and there are
+ * possibly other Shared copies in the memory system. A downstream
+ * cache helps in orchestrating the invalidation of these copies
+ * by sending out the appropriate express snoops.
+ */
+ void setResponderHadWritable()
+ {
+ assert(cacheResponding());
+ flags.set(RESPONDER_HAD_WRITABLE);
+ }
+ bool responderHadWritable() const
+ { return flags.isSet(RESPONDER_HAD_WRITABLE); }
+
void setSuppressFuncError() { flags.set(SUPPRESS_FUNC_ERROR); }
bool suppressFuncError() const { return flags.isSet(SUPPRESS_FUNC_ERROR); }
- void setPrefetchSquashed() { flags.set(PREFETCH_SNOOP_SQUASH); }
- bool prefetchSquashed() const { return flags.isSet(PREFETCH_SNOOP_SQUASH); }
+ void setBlockCached() { flags.set(BLOCK_CACHED); }
+ bool isBlockCached() const { return flags.isSet(BLOCK_CACHED); }
+ void clearBlockCached() { flags.clear(BLOCK_CACHED); }
// Network error conditions... encapsulate them as methods since
// their encoding keeps changing (from result field to command
cmd = MemCmd::BadAddressError;
}
- bool hadBadAddress() const { return cmd == MemCmd::BadAddressError; }
void copyError(Packet *pkt) { assert(pkt->isError()); cmd = pkt->cmd; }
- bool isSrcValid() const { return src != InvalidPortID; }
- /// Accessor function to get the source index of the packet.
- PortID getSrc() const { assert(isSrcValid()); return src; }
- /// Accessor function to set the source index of the packet.
- void setSrc(PortID _src) { src = _src; }
-
- bool isDestValid() const { return dest != InvalidPortID; }
- /// Accessor function for the destination index of the packet.
- PortID getDest() const { assert(isDestValid()); return dest; }
- /// Accessor function to set the destination index of the packet.
- void setDest(PortID _dest) { dest = _dest; }
- /// Reset destination field, e.g. to turn a response into a request again.
- void clearDest() { dest = InvalidPortID; }
-
Addr getAddr() const { assert(flags.isSet(VALID_ADDR)); return addr; }
/**
* Update the address of this packet mid-transaction. This is used
void setAddr(Addr _addr) { assert(flags.isSet(VALID_ADDR)); addr = _addr; }
unsigned getSize() const { assert(flags.isSet(VALID_SIZE)); return size; }
- Addr getOffset(int blkSize) const { return getAddr() & (Addr)(blkSize - 1); }
+
+ Addr getOffset(unsigned int blk_size) const
+ {
+ return getAddr() & Addr(blk_size - 1);
+ }
+
+ Addr getBlockAddr(unsigned int blk_size) const
+ {
+ return getAddr() & ~(Addr(blk_size - 1));
+ }
bool isSecure() const
{
return _isSecure;
}
+ /**
+ * Accessor function to atomic op.
+ */
+ AtomicOpFunctor *getAtomicOp() const { return req->getAtomicOpFunctor(); }
+ bool isAtomicOp() const { return req->isAtomic(); }
+
/**
* It has been determined that the SC packet should successfully update
- * memory. Therefore, convert this SC packet to a normal write.
+ * memory. Therefore, convert this SC packet to a normal write.
*/
void
convertScToWrite()
}
/**
- * When ruby is in use, Ruby will monitor the cache line and thus M5
- * phys memory should treat LL ops as normal reads.
+ * When ruby is in use, Ruby will monitor the cache line and the
+ * phys memory should treat LL ops as normal reads.
*/
void
convertLlToRead()
}
/**
- * Constructor. Note that a Request object must be constructed
+ * Constructor. Note that a Request object must be constructed
* first, but the Requests's physical address and size fields need
* not be valid. The command must be supplied.
*/
- Packet(Request *_req, MemCmd _cmd)
+ Packet(const RequestPtr _req, MemCmd _cmd)
: cmd(_cmd), req(_req), data(nullptr), addr(0), _isSecure(false),
- size(0), src(InvalidPortID), dest(InvalidPortID),
- bytesValidStart(0), bytesValidEnd(0),
- firstWordDelay(0), lastWordDelay(0),
+ size(0), headerDelay(0), snoopDelay(0), payloadDelay(0),
senderState(NULL)
{
if (req->hasPaddr()) {
* a request that is for a whole block, not the address from the
* req. this allows for overriding the size/addr of the req.
*/
- Packet(Request *_req, MemCmd _cmd, int _blkSize)
+ Packet(const RequestPtr _req, MemCmd _cmd, int _blkSize)
: cmd(_cmd), req(_req), data(nullptr), addr(0), _isSecure(false),
- src(InvalidPortID), dest(InvalidPortID),
- bytesValidStart(0), bytesValidEnd(0),
- firstWordDelay(0), lastWordDelay(0),
+ headerDelay(0), snoopDelay(0), payloadDelay(0),
senderState(NULL)
{
if (req->hasPaddr()) {
* less than that of the original packet. In this case the new
* packet should allocate its own data.
*/
- Packet(Packet *pkt, bool clearFlags = false)
+ Packet(const PacketPtr pkt, bool clear_flags, bool alloc_data)
: cmd(pkt->cmd), req(pkt->req),
- data(pkt->flags.isSet(STATIC_DATA) ? pkt->data : NULL),
+ data(nullptr),
addr(pkt->addr), _isSecure(pkt->_isSecure), size(pkt->size),
- src(pkt->src), dest(pkt->dest),
- bytesValidStart(pkt->bytesValidStart),
- bytesValidEnd(pkt->bytesValidEnd),
- firstWordDelay(pkt->firstWordDelay),
- lastWordDelay(pkt->lastWordDelay),
+ bytesValid(pkt->bytesValid),
+ headerDelay(pkt->headerDelay),
+ snoopDelay(0),
+ payloadDelay(pkt->payloadDelay),
senderState(pkt->senderState)
{
- if (!clearFlags)
+ if (!clear_flags)
flags.set(pkt->flags & COPY_FLAGS);
flags.set(pkt->flags & (VALID_ADDR|VALID_SIZE));
- flags.set(pkt->flags & STATIC_DATA);
+
+ // should we allocate space for data, or not, the express
+ // snoops do not need to carry any data as they only serve to
+ // co-ordinate state changes
+ if (alloc_data) {
+ // even if asked to allocate data, if the original packet
+ // holds static data, then the sender will not be doing
+ // any memcpy on receiving the response, thus we simply
+ // carry the pointer forward
+ if (pkt->flags.isSet(STATIC_DATA)) {
+ data = pkt->data;
+ flags.set(STATIC_DATA);
+ } else {
+ allocate();
+ }
+ }
}
/**
- * Change the packet type based on request type.
+ * Generate the appropriate read MemCmd based on the Request flags.
*/
- void
- refineCommand()
+ static MemCmd
+ makeReadCmd(const RequestPtr req)
{
- if (cmd == MemCmd::ReadReq) {
- if (req->isLLSC()) {
- cmd = MemCmd::LoadLockedReq;
- } else if (req->isPrefetch()) {
- cmd = MemCmd::SoftPFReq;
- }
- } else if (cmd == MemCmd::WriteReq) {
- if (req->isLLSC()) {
- cmd = MemCmd::StoreCondReq;
- } else if (req->isSwap()) {
- cmd = MemCmd::SwapReq;
- }
- }
+ if (req->isLLSC())
+ return MemCmd::LoadLockedReq;
+ else if (req->isPrefetch())
+ return MemCmd::SoftPFReq;
+ else
+ return MemCmd::ReadReq;
+ }
+
+ /**
+ * Generate the appropriate write MemCmd based on the Request flags.
+ */
+ static MemCmd
+ makeWriteCmd(const RequestPtr req)
+ {
+ if (req->isLLSC())
+ return MemCmd::StoreCondReq;
+ else if (req->isSwap())
+ return MemCmd::SwapReq;
+ else
+ return MemCmd::WriteReq;
}
/**
* Constructor-like methods that return Packets based on Request objects.
- * Will call refineCommand() to fine-tune the Packet type if it's not a
- * vanilla read or write.
+ * Fine-tune the MemCmd type if it's not a vanilla read or write.
*/
static PacketPtr
- createRead(Request *req)
+ createRead(const RequestPtr req)
{
- PacketPtr pkt = new Packet(req, MemCmd::ReadReq);
- pkt->refineCommand();
- return pkt;
+ return new Packet(req, makeReadCmd(req));
}
static PacketPtr
- createWrite(Request *req)
+ createWrite(const RequestPtr req)
{
- PacketPtr pkt = new Packet(req, MemCmd::WriteReq);
- pkt->refineCommand();
- return pkt;
+ return new Packet(req, makeWriteCmd(req));
}
/**
*/
~Packet()
{
- // If this is a request packet for which there's no response,
- // delete the request object here, since the requester will
- // never get the chance.
- if (req && isRequest() && !needsResponse())
+ // Delete the request object if this is a request packet which
+ // does not need a response, because the requester will not get
+ // a chance. If the request packet needs a response then the
+ // request will be deleted on receipt of the response
+ // packet. We also make sure to never delete the request for
+ // express snoops, even for cases when responses are not
+ // needed (CleanEvict and Writeback), since the snoop packet
+ // re-uses the same request.
+ if (req && isRequest() && !needsResponse() &&
+ !isExpressSnoop()) {
delete req;
- deleteData();
- }
-
- /**
- * Reinitialize packet address and size from the associated
- * Request object, and reset other fields that may have been
- * modified by a previous transaction. Typically called when a
- * statically allocated Request/Packet pair is reused for multiple
- * transactions.
- */
- void
- reinitFromRequest()
- {
- assert(req->hasPaddr());
- flags = 0;
- addr = req->getPaddr();
- _isSecure = req->isSecure();
- size = req->getSize();
-
- src = InvalidPortID;
- dest = InvalidPortID;
- bytesValidStart = 0;
- bytesValidEnd = 0;
- firstWordDelay = 0;
- lastWordDelay = 0;
-
- flags.set(VALID_ADDR|VALID_SIZE);
+ }
deleteData();
}
/**
* Take a request packet and modify it in place to be suitable for
- * returning as a response to that request. The source field is
- * turned into the destination, and subsequently cleared. Note
- * that the latter is not necessary for atomic requests, but
- * causes no harm as neither field is valid.
+ * returning as a response to that request.
*/
void
makeResponse()
{
assert(needsResponse());
assert(isRequest());
- origCmd = cmd;
cmd = cmd.responseCommand();
// responses are never express, even if the snoop that
// triggered them was
flags.clear(EXPRESS_SNOOP);
-
- dest = src;
- src = InvalidPortID;
}
void
}
+ public:
+ /**
+ * @{
+ * @name Data accessor mehtods
+ */
+
/**
* Set the data pointer to the following value that should not be
- * freed.
+ * freed. Static data allows us to do a single memcpy even if
+ * multiple packets are required to get from source to destination
+ * and back. In essence the pointer is set calling dataStatic on
+ * the original packet, and whenever this packet is copied and
+ * forwarded the same pointer is passed on. When a packet
+ * eventually reaches the destination holding the data, it is
+ * copied once into the location originally set. On the way back
+ * to the source, no copies are necessary.
*/
template <typename T>
void
dataStatic(T *p)
{
- assert(flags.noneSet(STATIC_DATA|DYNAMIC_DATA|ARRAY_DATA));
+ assert(flags.noneSet(STATIC_DATA|DYNAMIC_DATA));
data = (PacketDataPtr)p;
flags.set(STATIC_DATA);
}
/**
- * Set the data pointer to a value that should have delete []
- * called on it.
+ * Set the data pointer to the following value that should not be
+ * freed. This version of the function allows the pointer passed
+ * to us to be const. To avoid issues down the line we cast the
+ * constness away, the alternative would be to keep both a const
+ * and non-const data pointer and cleverly choose between
+ * them. Note that this is only allowed for static data.
*/
template <typename T>
void
- dataDynamicArray(T *p)
+ dataStaticConst(const T *p)
{
- assert(flags.noneSet(STATIC_DATA|DYNAMIC_DATA|ARRAY_DATA));
- data = (PacketDataPtr)p;
- flags.set(DYNAMIC_DATA|ARRAY_DATA);
+ assert(flags.noneSet(STATIC_DATA|DYNAMIC_DATA));
+ data = const_cast<PacketDataPtr>(p);
+ flags.set(STATIC_DATA);
}
/**
- * set the data pointer to a value that should have delete called
- * on it.
+ * Set the data pointer to a value that should have delete []
+ * called on it. Dynamic data is local to this packet, and as the
+ * packet travels from source to destination, forwarded packets
+ * will allocate their own data. When a packet reaches the final
+ * destination it will populate the dynamic data of that specific
+ * packet, and on the way back towards the source, memcpy will be
+ * invoked in every step where a new packet was created e.g. in
+ * the caches. Ultimately when the response reaches the source a
+ * final memcpy is needed to extract the data from the packet
+ * before it is deallocated.
*/
template <typename T>
void
dataDynamic(T *p)
{
- assert(flags.noneSet(STATIC_DATA|DYNAMIC_DATA|ARRAY_DATA));
+ assert(flags.noneSet(STATIC_DATA|DYNAMIC_DATA));
data = (PacketDataPtr)p;
flags.set(DYNAMIC_DATA);
}
}
/**
- * return the value of what is pointed to in the packet.
+ * Get the data in the packet byte swapped from big endian to
+ * host endian.
+ */
+ template <typename T>
+ T getBE() const;
+
+ /**
+ * Get the data in the packet byte swapped from little endian to
+ * host endian.
+ */
+ template <typename T>
+ T getLE() const;
+
+ /**
+ * Get the data in the packet byte swapped from the specified
+ * endianness.
+ */
+ template <typename T>
+ T get(ByteOrder endian) const;
+
+ /**
+ * Get the data in the packet byte swapped from guest to host
+ * endian.
*/
template <typename T>
T get() const;
+ /** Set the value in the data pointer to v as big endian. */
+ template <typename T>
+ void setBE(T v);
+
+ /** Set the value in the data pointer to v as little endian. */
+ template <typename T>
+ void setLE(T v);
+
/**
- * set the value in the data pointer to v.
+ * Set the value in the data pointer to v using the specified
+ * endianness.
*/
template <typename T>
+ void set(T v, ByteOrder endian);
+
+ /** Set the value in the data pointer to v as guest endian. */
+ template <typename T>
void set(T v);
/**
void
setData(const uint8_t *p)
{
+ // we should never be copying data onto itself, which means we
+ // must idenfity packets with static data, as they carry the
+ // same pointer from source to destination and back
+ assert(p != getPtr<uint8_t>() || flags.isSet(STATIC_DATA));
+
if (p != getPtr<uint8_t>())
+ // for packet with allocated dynamic data, we copy data from
+ // one to the other, e.g. a forwarded response to a response
std::memcpy(getPtr<uint8_t>(), p, getSize());
}
void
deleteData()
{
- if (flags.isSet(ARRAY_DATA))
+ if (flags.isSet(DYNAMIC_DATA))
delete [] data;
- else if (flags.isSet(DYNAMIC_DATA))
- delete data;
- flags.clear(STATIC_DATA|DYNAMIC_DATA|ARRAY_DATA);
+ flags.clear(STATIC_DATA|DYNAMIC_DATA);
data = NULL;
}
- /** If there isn't data in the packet, allocate some. */
+ /** Allocate memory for the packet. */
void
allocate()
{
- if (data) {
- assert(flags.isSet(STATIC_DATA|DYNAMIC_DATA));
- return;
+ // if either this command or the response command has a data
+ // payload, actually allocate space
+ if (hasData() || hasRespData()) {
+ assert(flags.noneSet(STATIC_DATA|DYNAMIC_DATA));
+ flags.set(DYNAMIC_DATA);
+ data = new uint8_t[getSize()];
}
-
- assert(flags.noneSet(STATIC_DATA|DYNAMIC_DATA));
- flags.set(DYNAMIC_DATA|ARRAY_DATA);
- data = new uint8_t[getSize()];
}
- /**
- * Check a functional request against a memory value represented
- * by a base/size pair and an associated data array. If the
- * functional request is a read, it may be satisfied by the memory
- * value. If the functional request is a write, it may update the
- * memory value.
- */
- bool checkFunctional(Printable *obj, Addr base, bool is_secure, int size,
- uint8_t *data);
+ /** @} */
+
+ private: // Private data accessor methods
+ /** Get the data in the packet without byte swapping. */
+ template <typename T>
+ T getRaw() const;
+
+ /** Set the value in the data pointer to v without byte swapping. */
+ template <typename T>
+ void setRaw(T v);
+ public:
/**
* Check a functional request against a memory value stored in
- * another packet (i.e. an in-transit request or response).
+ * another packet (i.e. an in-transit request or
+ * response). Returns true if the current packet is a read, and
+ * the other packet provides the data, which is then copied to the
+ * current packet. If the current packet is a write, and the other
+ * packet intersects this one, then we update the data
+ * accordingly.
*/
bool
- checkFunctional(PacketPtr other)
+ checkFunctional(PacketPtr other)
{
- uint8_t *data = other->hasData() ? other->getPtr<uint8_t>() : NULL;
+ // all packets that are carrying a payload should have a valid
+ // data pointer
return checkFunctional(other, other->getAddr(), other->isSecure(),
- other->getSize(), data);
+ other->getSize(),
+ other->hasData() ?
+ other->getPtr<uint8_t>() : NULL);
}
+ /**
+ * Does the request need to check for cached copies of the same block
+ * in the memory hierarchy above.
+ **/
+ bool
+ mustCheckAbove() const
+ {
+ return cmd == MemCmd::HardPFReq || isEviction();
+ }
+
+ /**
+ * Is this packet a clean eviction, including both actual clean
+ * evict packets, but also clean writebacks.
+ */
+ bool
+ isCleanEviction() const
+ {
+ return cmd == MemCmd::CleanEvict || cmd == MemCmd::WritebackClean;
+ }
+
+ /**
+ * Check a functional request against a memory value represented
+ * by a base/size pair and an associated data array. If the
+ * current packet is a read, it may be satisfied by the memory
+ * value. If the current packet is a write, it may update the
+ * memory value.
+ */
+ bool
+ checkFunctional(Printable *obj, Addr base, bool is_secure, int size,
+ uint8_t *_data);
+
/**
* Push label for PrintReq (safe to call unconditionally).
*/